GeForce DDR slower than GF2 MX?!

[ahfung]

I always thought that GF DDR faster than GF2 MX but it seems I'm wrong. :Q

The tech demo Crytek X ISLE runs way faster with MX than DDR under "realtime environment cubemapping in refraction/reflection mode". The framerate drops to around 12fps for an overclocked GF DDR, while a GF2 MX running at spec speed gets over 30fps. My GF DDR would only get about 12fps independent of resolution and color depth, overclocked or not. I'm in great doubt now. What accounts for the difference?

Take a look at the pics and notice the current fps:

Environment cubemapping in refraction mode
Environment cubemapping in reflection mode

 

[Packet]

Hmm, the DDR should be faster

Never used the Crytek tech demo before, does it use T&L?
Remember that the MX has a more advanced T&L than the original geforce line.

 

[Smbu]

Doesn't the MX have a higher fillrate compared to the Original Geforce 256 cards? Plus the MX can do 2 texels per pipeline on it's 2 pipelines, while the 256 cards can only do 1 texel per pipeline, but they do have 4 pipelines.

 

[ahfung]

Remember that the MX has a more advanced T&L than the original geforce line.

I know, MX core is running at 175MHz while classic GeForce at 120MHz. They have the same T&L engine but MX's is 55MHz faster. But this still doesn't account for the difference of 12 vs > 30fps, not to mentioned my GF DDR is overclocked to 150/350.

Doesn't the MX have a higher fillrate compared to the Original Geforce 256 cards?

Yes MX has higher multitexture fillrate, BUT this only happens at 640x480 and at anything above MX will be severly crippled by its memory.

 

[Smbu]

The MX can still hold up pretty well in higher resolutions if it is doing it in 16bit color(so the memory bandwidth issue doesn't affect it as much). Just because the DDR card has all of that extra memory bandwidth doesn't mean it can use it, meaning the card could be fillrate limited and not bandwidth limited. It's just like when you OC the memory on GTS, Pro, and Ultra cards. You don't see the benefit of the increased bandwidth until you go into the higher resolutions(1024x768x32bpp and higher).

 

[ahfung]

MX is already no contest with DDR under Q3A HQ 800x600 or default 3D Mark 2000 (1024x768x16)

If you benchmark them with 3D Mark 2000 - Fillrate tests, you'd find that at any resolution above 640x480x16 MX is memory bandwidth limited.